sync: master → develop 2026-06-14 — pull #546 release + auto-version bumps by rsalcara · Pull Request #549 · rsalcara/InfiniteAPI

rsalcara · 2026-06-14T15:00:05Z

Resumo

Sync periódico master → develop após o merge da PR #546. Develop estava 27 commits atrás. Merge clean, sem conflitos.

O que vem nessa sync (27 commits)

Audit fixes da release #546 (8)

`7fdf1898b7` chore: merge master version bumps into release branch
`8b0176b8fa` fix(PR release: develop → master 2026-06-13 — VoIP + 6 audit batches + version sync #546 audit): tr range BLOCKER + 5 VoIP P1s + JID-HOSTED-P2 + 4 P2/P3 polish
`53cbf6399c` fix(audit r6): FU-04 varint 5-byte overflow + FU-06 port upper-bound + FU-03 reverse-delete forward cleanup
`6303c2dd10` release: develop → master 2026-06-13 (merge commit)

Auto-bot version bumps (16)

PRs #516, #517, #524, #525, #527, #528, #530, #531, #539, #540, #541, #542, #543, #544, #547, #548 — atualizações automáticas de `baileys-version.json` e `WAProto/*`.

Release commits anteriores (3)

`bb614011c6` release: develop → master 2026-06-08 (release chore(build): exclude .md from src→lib mirror (keep docs repo-local) #514+chore(logs): compact recoverable Signal Protocol failures #515+feat(meta-ai): decrypt msmsg via empirically-validated WA Web algorithm #518+feat(stickers): support lottieStickerMessage for animated .was stickers #519+fix(decode): preserve outer messageContextInfo when unwrapping deviceSentMessage #520)
`285db5914b` release: develop → master 2026-06-08
`424260a4e2` release: develop → master 2026-06-06 (v2)

Estado pós-sync

Item	Antes	Depois
`baileys-version.json`	`[2,3000,1041103517]`	`[2,3000,1041437765]` ✅
`update-proto.yml` `tr` bug	presente	corrigido ✅
Audit fixes round 5/6	ausentes	presentes ✅
Diff develop ↔ master	27 atrás	0 (igualados)

Customizações intactas

Carrossel, lista, botão, poll, view-once, biz quality_control, Lottie wrap, Meta AI msmsg, DSM, TC token, LID↔PN, Phase 9 multi-DB, useLegacyLock, schema migrations — todos preservados.

🤖 Generated with Claude Code

Summary by cubic

Sync master → develop after the release to align versions and bring audit fixes that improve VoIP reliability, JID @hosted.lid handling, Signal varint parsing, env/metrics robustness, and the proto auto-update workflow.

Bug Fixes
- CI: fix tr character class in .github/workflows/update-proto.yml to stop Summary step failures.
- VoIP: bail when any destination encryption fails; detach CB:call/CB:receipt listeners on disconnect; correctly handle @hosted.lid; re-register global callbacks on init, surface init errors, and add a 30s worker WASM load timeout.
- Signal: reject 5-byte varints that overflow 32-bit in libsignal.ts.
- Env/Metrics: intFromEnv/floatFromEnv trim input; intFromEnv adds max and clamps Prometheus port to 1–65535.
- Multi-DB SQLite: reverse-delete also clears legacy forward entries; cache statement for getAllLidsForPn; clarify error handling comments.
Dependencies
- Update WAProto/* and src/Defaults/baileys-version.json to the latest WhatsApp Web/proto versions (WAProto.proto header and baileys-version.json now at 2.3000.1041437765).

^{Written for commit 91df359. Summary will update on new commits.}

…f path (#513 review) PR #513 chatgpt-codex review (P2) caught a real gap that PR #509 introduced when it added the `maxKeys` cap to `userDevicesCache`: - PR #509 added `safeCacheSet` handling to the cache-population path in `messages-send.ts:528-548` (where USync fills the cache). - The OTHER call-site that writes to `userDevicesCache` — `messages-recv.ts:2240` in the device-notification handler ('add' / 'remove' tags) — was left unprotected. `@cacheable/node-cache`'s capacity check is `keyCount() + 1 > maxKeys` — applied BEFORE the engine checks whether the key already exists. So once the cache reaches its `maxKeys` ceiling (5,000 entries), even an UPDATE to an already-cached user throws ECACHEFULL. The guard `if (!existingCache.length) continue` at line 2210 doesn't help: it only short-circuits when the user is NOT cached. A device-list update for an already-cached user would still hit the `.set` and throw. Real impact under sustained gateway load: - Cache hits 5,000 entries (a few dozen active groups will get there). - A device-add/remove notification arrives for one of the cached users. - `userDevicesCache.set(...)` throws ECACHEFULL. - The throw propagates into the message-receive handler and lands in Baileys' outer error boundary — connection survives, but a log-level error is emitted and the affected user's device list is NOT updated in the cache. - Next message-send for that user fetches the fresh list via USync (`getUSyncDevices`), which IS guarded — so the durable behavior eventually recovers. The fix routes the device-notif write through the same `safeCacheSet` helper PR #509 used in messages-send.ts. Same swallow-ECACHEFULL semantics, same debug log, same `getUSyncDevices` fallback path. Symmetric handling across the two call-sites that update the cache. Test plan: - npm run build ✓ (3 phases pass, zero errors) - 1 line of production change in `messages-recv.ts:2240`, plus comment explaining the WHY. No test changes needed — `safeCacheSet`'s behavior is already covered by `cache-utils.test.ts`. Out of scope (intentionally NOT touched): - Carousel, lists, buttons, polls, view-once, biz quality_control, useLegacyLock, TC token custom flow, LID↔PN batched, Phase 9 multi-DB, lidDbMigrated:false, the cacheMetricsInterval memory-leak fix, schema migrations + statement cache + busy retry. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…#513) release: develop → master 2026-06-06 (v2) — #509 + #510 + #511 + #512 (#513)

Co-authored-by: rsalcara <rsalcara@users.noreply.github.com>

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Merges 6 PRs landed on `develop` since the previous master sync (2026-06-06), preserves the two auto-bumps that have since landed on master (#516, #517), and ships them together. PRs included in this release ============================ * #514 — chore(build): exclude .md from src→lib mirror Internal architecture notes (`src/Utils/multi-db-sqlite/ARCHITECTURE.md`, `USAGE.md`) no longer ship in the published package. Disk savings on consumer side + stops leaking internal phase planning into anyone that runs `npm install`. * #515 — chore(logs): compact recoverable Signal Protocol failures Recoverable Signal failures (Bad MAC, MessageCounterError, old counter, Invalid PreKey ID) are now compact one-liners at `warn` instead of three multi-KB `error` entries each. Per-event log output: ~10KB → ~250B. Unknown errors keep full stack + stanza. Net: error-dashboard noise drops dramatically without losing real signal. * #518 — feat(meta-ai): decrypt msmsg via empirically-validated WA Web Meta AI / FBID-bot replies (`<enc type="msmsg">`) now decrypt instead of NACKing with MissingMessageSecret. Algorithm reverse-engineered from `WAWebBotMessageSecret` via CDP and validated LIVE with a Meta AI "oi" capture. Five distinct bugs in upstream PR WhiskeySockets#2592 fixed (unbounded cache, cross-tenant leak, no FBID dispatch, drops streaming chunks, raw-id cache key, 12-strat brute force). Per-socket bounded LRU + `flushAll() + close()` on socket end to release the NodeCache timer. * #519 — feat(stickers): support lottieStickerMessage for animated .was Wraps + unwraps Lottie stickers (`application/was`) in `lottieStickerMessage` (FutureProofMessage at proto field 74), matching `WAWebE2EProtoGenerator`'s actual implementation. Mobile clients silently drop Lottie stickers in the unwrapped form; this makes them render on Android/iOS. Send-path `getMediaType` + DSM `messageContextInfo` reads both walk the wrap correctly. * #520 — fix(decode): preserve outer messageContextInfo when unwrapping deviceSentMessage Closes the codex P1 gap left deferred from #518. When WhatsApp delivers a fromMe-via-linked-device message, the OUTER `Message` carries `messageContextInfo.messageSecret` and the inner does not — the previous `msg = msg.deviceSentMessage?.message || msg` dropped that secret entirely. New `unwrapDeviceSentMessage` does a per-field merge matching WA Web's `l(e)` source exactly, including a length-check guard on `threadId` (protobuf decodes empty `repeated` as `[]` not `undefined` — naive `??` would lose the outer thread context). * #513 follow-up (already on master as `373a6f3622` from previous release) — the develop-side commit `50f9a77ef8` is the same change. Merge collapses both onto the master line; no behaviour change. Preserved from master (NOT touched by this merge) ================================================== * #516 — chore: update proto/version to v2.3000.1040989513 Auto-bumped `WAProto/WAProto.proto` + `baileys-version.json`. develop never touched these files, so the bump survives intact. * #517 — chore: update WhatsApp Web version to v2.3000.1040994085 Auto-bumped `baileys-version.json` again. Same — develop didn't touch it, master's value (`{"version":[2,3000,1040994085]}`) is preserved verbatim post-merge. Validation ========== * Auto-merge succeeded with ZERO conflicts — `messages-recv.ts` was the only file modified on both sides, but the edits were non-overlapping and git resolved them automatically. * `npm run build` clean on the merged tree. * 57/57 tests pass across the new-PR suites (dsm-context-info + meta-ai-msmsg + lottie-sticker-message + error-log-utils + process-message.protocol-guard). * `baileys-version.json` post-merge: `{"version":[2,3000,1040994085]}` — confirms master's auto-bump was preserved (NOT downgraded to develop's stale `1040973525`). Customizations preserved (NOT touched) ====================================== Carrossel, lists, buttons, polls, view-once, biz `quality_control`, `useLegacyLock`, TC token custom flow, LID↔PN batched, Phase 9 multi-DB, `lidDbMigrated:false`, `cacheMetricsInterval` memory-leak fix, schema migrations + statement cache + busy retry. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

5 findings from the release PR #521 audit applied before promoting develop → master. The release branch itself doesn't change; this PR lands the fixes on develop and the release branch will be refreshed to include them. P2 — Thread 1 (chatgpt) — reporting-token check reads outer wrap ================================================================ `messages-send.ts:1884` gated reporting-token attachment on `reportingMessage?.messageContextInfo?.messageSecret`. For Lottie stickers post-PR #519 that secret lives INSIDE the `lottieStickerMessage` wrap — the top-level read returns undefined and the token is silently skipped. Regression vs plain sticker behaviour. Fix: lift `messageContextInfo` from both locations the same way the initial-fanout DSM read does (the lift PR #519 already shipped on line ~1198): const reportingMessageSecret = reportingMessage?.lottieStickerMessage?.message ?.messageContextInfo?.messageSecret ?? reportingMessage?.messageContextInfo?.messageSecret Required adding `reportingMessage &&` to the surrounding `if` so TypeScript doesn't lose the narrowing it previously got from the inline property chain (the narrowing was the side-effect of reading `reportingMessage.messageContextInfo` in the condition itself). P2 — Thread 5 (coderabbit Major) — retry-resend DSM drops messageContextInfo ============================================================================ `messages-send.ts:1580` rebuilds the `deviceSentMessage` envelope on the retry-resend path WITHOUT carrying `messageContextInfo` along. The initial-fanout path got the lift in PR #519 (line ~1198); the retry path was missed. Concrete impact: when a companion device receives the RETRY of a Lottie sticker via DSM, our own `unwrapDeviceSentMessage` finds: - inner = messageToSend = { lottieStickerMessage: { message: { stickerMessage, messageContextInfo } } } - inner.messageContextInfo = undefined (it's nested in the wrap) - outer.messageContextInfo = undefined (the retry envelope omitted it) → messageSecret = undefined, reporting token / encrypted-edit decryption material lost on the companion's copy. Fix: same lift inline at the retry envelope build site. P2 — Thread 2 (chatgpt) — OrphanMsmsgError stub goes to Signal retry path ========================================================================= When `decryptMsmsgBotMessage` raises `OrphanMsmsgError` (cache miss), the catch in `decode-wa-message.ts` sets `messageStubType = CIPHERTEXT` and `messageStubParameters[0] = String(err.message)`. The receive handler at `messages-recv.ts:3115` checks for known stub-param strings and falls through to the Signal retry / PDO placeholder-resend path when none match — burning retry budget asking the bot for prekeys it has no business issuing, for a problem (missing CACHE entry) that a Signal retry can never fix. Fix: add a guard for `messageStubParameters[0]?.startsWith('decryptMsmsgBotMessage:')` right after the `MISSING_KEYS_ERROR_TEXT` branch. Plain ACK (no NACK, no retry) so the server considers the message delivered. The next bot reply that arrives after the outgoing-secret cache populates will decrypt cleanly. NOT a NACK MissingMessageSecret: that would tell the server to retransmit, and the retransmission will hit the same orphan state until the outgoing-side cache is populated (deferred per PR #518). P3 — Thread 4 (copilot) — `__internal` export not consumed ========================================================== `src/Utils/meta-ai-msmsg.ts` exported an `__internal` bag of helpers (`BOT_MESSAGE_INFO`, `KEY_LENGTH`, `isMeJid`, `deriveKeyAndDecrypt`, `decodeDecryptedMsmsg`, `userOnlyJid`, `isJidGroup`) that no file in the repo actually imports. Grep confirms zero call sites. Removed — the helpers stay module-private as intended. P3 — Thread 3 (copilot) — `(err as any)?.message` in auth-utils =============================================================== `src/Utils/auth-utils.ts:714` (the trace I added in PR #515 to log `transactWith rolled back` without duplicating the stack) used a plain `(err as any)?.message` cast. Tightened to `err instanceof Error ? err.message : String(err)` — no runtime change, just removes the `any` cast and gives string fallback for non-Error throws. Validation ========== * `npm run build` clean (TS narrowing fixed via explicit `reportingMessage &&` guard). * 57/57 tests pass across the new-PR suites (dsm-context-info-preservation + meta-ai-msmsg + lottie-sticker-message + error-log-utils + process-message.protocol-guard). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Two findings from the cubic re-review on PR #521 that the previous audit-fix commit (cherry-picked from closed PR #522) didn't cover. Issue D (P2) — test re-implemented production helper ==================================================== cubic thread 8: `dsm-context-info-preservation.test.ts` kept a local copy of `unwrapDeviceSentMessage` and asserted against it. Comment claimed "if rules change in decode-wa-message.ts these break first" but that was wrong — the assertions tested the LOCAL copy, not production. Reverting `decode-wa-message.ts` to the old `msg = msg.deviceSentMessage?.message || msg` would leave all 10 tests green. The tests were validating themselves. Fix: * `decode-wa-message.ts` — flipped the helper from module-private `const` to `export const unwrapDeviceSentMessage`. The function is a leaf utility; exporting it surfaces nothing operational beyond what tests need. * `dsm-context-info-preservation.test.ts` — deleted the re-derived copy, imported from `../../Utils/decode-wa-message`. Header comment updated to explain WHY we now import (production parity) instead of re-derive. Issue E (P3) — `if (!err)` too broad in compactError ==================================================== cubic thread 9: `error-log-utils.ts:33` short-circuited every falsy input (including `0`, `''`, `false`, `NaN`) to the literal string `'Unknown'`. Those are unusual but valid thrown values — code that does `throw 0` would get its actual value erased. Operational impact near zero (Signal Protocol throws `Error` instances), but the contract is wrong. Fix: * `error-log-utils.ts` — `if (!err)` → `if (err == null)`. Comment added explaining the nullish-only narrowing. * `error-log-utils.test.ts` — new test `preserves falsy primitives that are NOT null/undefined` pins the new contract: `compactError(0)` → `'0'`, `compactError(false)` → `'false'`, etc. Validation ========== * `npm run build` clean * 58/58 tests pass (5 suites: dsm-context-info + error-log-utils + meta-ai-msmsg + lottie-sticker-message + process-message). Was 57; +1 from the new falsy-primitives test. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

CI lint failed on the previous commit: - error-log-utils.ts:36 — `if (err == null)` violated the codebase's `eqeqeq` rule even though `== null` is the idiomatic nullish check. Expanded to `if (err === null || err === undefined)` for the same semantics with strict-equality compliance. - meta-ai-msmsg.ts:46 — `isJidGroup` was imported but its only consumer (the now-removed `__internal` export bag from cubic audit thread 4) is gone. Removed from the import list. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

cubic re-review (thread 13, confidence 9) caught that the previous ACK guard was too broad: if (msg?.messageStubParameters?.[0]?.startsWith('decryptMsmsgBotMessage:')) `decryptMsmsgBotMessage` throws with a `decryptMsmsgBotMessage:` prefix for several distinct conditions, not just `OrphanMsmsgError`: - `'decryptMsmsgBotMessage: no messageSecret for ${cacheKey}'` ← OrphanMsmsgError — cache miss, ACK + wait for cache to populate - `'decryptMsmsgBotMessage: missing meta.target_id'` ← malformed stanza — deserves NACK/retry - `'decryptMsmsgBotMessage: MessageSecretMessage missing encIv/encPayload'` ← malformed proto — deserves NACK/retry - real AES-GCM auth-tag mismatch (string varies but may also start with `decryptMsmsgBotMessage:` if wrapped) ← deserves the Signal retry path The broader prefix silently ACK'd all of those, hiding real protocol failures from the server-side retry machinery. Narrow the match to the exact substring that uniquely identifies the orphan-cache case: startsWith('decryptMsmsgBotMessage: no messageSecret for ') Other `decryptMsmsgBotMessage:` failures now flow normally to the Signal retry / PDO placeholder-resend path (where they belong). Comment expanded to document why the narrow match is required. Validation: * build clean * 58/58 tests still pass Thread 12 (P3) — `String(err)` in auth-utils.ts:714 — is cosmetic and deferred per the audit recommendation. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

release: develop → master 2026-06-08 (5 PRs)